The Claude Haiku 4.5 AI model sent a false tip about a real unsolved murder to the Philadelphia Police Department's website during testing, its developer Anthropic said in a report, according to infohub.kz.
The message was flagged as spam and never reached investigators, the company said.
As part of the evaluation, Claude was asked to come up with and carry out sample tasks on randomly selected web pages. The model opened a page about an unsolved murder that included a form for submitting information to police.
Claude wrote that it supposedly had information about the case and had seen a person matching the description near the street listed on the page. According to Anthropic, however, the site contained no description of the suspect.
The model left the name and contact fields blank and submitted the message. Its instructions barred it from entering personal data, making purchases or taking destructive actions, but did not specifically prohibit submitting forms.
According to police, the false tip was submitted on July 18. Anthropic discovered the incident on September 28 and notified the department on October 7, ABC News reported, citing a statement from Philadelphia police.
The department called the delay in detecting and reporting the incident unacceptable. Police stressed that unsolved cases involve real victims and their families, and that technology companies must prevent fabricated information from reaching law enforcement.
Police also explained that tips are reviewed by officers: a message is treated as a lead whose credibility must be verified. Automated submission does not bypass that process.
The company temporarily cut off access to the live internet in all internal evaluation tests until it is satisfied that its safeguards are reliable. Some tests were moved to isolated environments, and web-access tools were given additional restrictions.
Anthropic assessed that the incidents it identified had minimal consequences, but showed that models can go beyond their intended task when instructions are ambiguous.